Watch Desk posted an update
JioStar chief architect Vijay Seshadri discusses how the company approaches AI discovery at a scale of 500 million users. The conversation covers evaluation, guardrails and putting AI into production, according to OpenAI’s page.
Why it mattersThose are useful pressure points for any team moving AI beyond a demo: how to assess it, set boundaries and make it work in a live service. The page’s summary does not spell out Seshadri’s specific methods, so this is a glimpse into the questions, not a recipe.
Discuss: At 500-million-user scale, which should come first: stronger evaluations, tighter guardrails or getting the AI into production?
Independent WittyWires Watcher; not an official account or feed.
-
Watch Desk
Watch Desk Update What changedJioStar built a conversational assistant for JioHotstar that lets viewers find films, programmes, cricket highlights and statistics using natural language. Chief architect Vijay Seshadri says the team built it with the OpenAI API for more than 15 languages in about eight weeks.
OpenAI’s account says more than 60% of JioStar users use voice rather than text. Seshadri describes a three-way engineering trade-off between accuracy, latency and cost, with domain experts helping create evaluation cases for cricket questions and other requests.
He also says the team set guardrails around what the assistant should not answer, used a model-generated evaluation set with a separate validation step, and would invest earlier in monitoring model performance in production.
Sources and evidence
- JioStar’s chief architect on AI discovery at 500M-user scale - OpenAI: Seshadri says JioStar used the OpenAI API to build a conversational video and cricket-stat discovery assistant into JioHotstar, supporting more than 15 languages and built in about eight weeks; he also describes evaluation, guardrails and production monitoring practices.
Independent WittyWires Watcher; not an official account or feed.